[论文] [论文] One Basis to Animate Them All: Gaussian Blendshape Distillation f...

论文概要

研究领域: CV 作者: Ramazan Fazylov, Stamatis Lefkimmiatis, Ivan Laptev 发布时间: 2026-10-01 arXiv: 2610.02207

中文摘要

3D 高斯化身支持快速渲染，但其实时动画常受高昂神经推理成本的制约。我们解决这一瓶颈，证明预训练化身模型的动画可以被一组与身份无关的 blendshape 线性组合紧密逼近。基于此发现，我们提出 GALA——通过线性逼近进行高斯动画的蒸馏方法，用浅层系数预测器加线性混合替代逐帧的重型神经解码。为提高保真度并降低内存需求，我们在渲染感知度量下按内存预算通过块局部 PCA 构建基。我们的方法学习一个浅层 MLP 网络预测 blendshape 系数，无需重训原始模型即可应用于多种动画架构。我们在三个不同的化身上验证了 GALA，涵盖面部表情和全身衣物动力学的 3D 动画。蒸馏结果可泛化到未见身份，CPU 动画成本降低达三个数量级，同时保持了大部分渲染质量。结果证实学习到的化身表征具有共享的线性结构，实现了高达 60fps 的移动设备端高效精确动画。

原文摘要

3D Gaussian avatars support fast rendering, however, their real-time animation is often challenged by the costly neural inference. We address this bottleneck and show that the animation of pretrained avatar models can be closely approximated by a linear combination of identity-independent blendshapes. Building on this finding, we introduce GALA (Gaussian Animation via Linear Approximation), a distillation method that replaces per-frame heavy neural decoding with a shallow coefficient predictor and a linear blend. To improve fidelity and reduce memory requirements, we propose to construct the basis using block-local PCA under a rendering-aware metric and a memory budget. Our method learns a shallow MLP network to predict blendshape coefficients and applies to various animation architectures...

#论文 #arXiv #CV #小凯